I’ve spent the last five years in recruiting, mostly headhunting senior engineers and ML leaders, and I can tell you the one sentence I have never once heard from a hiring manager: “we invented a new job.”
Nobody says it. What they say is “Dave quit.” Or “Rhea got promoted, we need to backfill.” Or, on a bad Monday, “we let the team go, can you refill it by Friday.”
After five years I finally asked myself a question I should have asked in year one. What is this thing we call the job market, actually? Because from where I sit it doesn’t look like a market in jobs. It looks like a market in seats.
The priest who saw it first
The answer was found in 1970, by a sociologist at Harvard named Harrison White, in a place about as far from tech recruiting as you can get: parish registers.
White had gotten hold of five decades of records from three American church bodies. Every assignment, every posting, every name and date of which priest held which parish and when. And he did the thing nobody had thought to do. He ignored the men completely and followed the jobs.
Here’s how it works. A bishop dies. His parish opens. A younger priest moves into it, which opens his parish, which pulls a third priest, which opens a fourth post. On paper, three men got promoted that year. Follow the actual mechanism and one vacancy rolled downhill, and the men were just the ball bearings.
That’s it. That’s the whole machine, and the machine has three moving parts. Jobs come in discrete units. One person holds one at a time. And you cannot take a seat that is occupied; someone has to leave first. Granovetter said it a few years later without the poetry: most job movement is not the creation of jobs but the transfer of incumbents between them.
Now the part of the story I refuse to believe anyone planned. The same mechanism runs hermit crabs.
Ivan Chase spent a decade of summers at a Long Island tide pool dropping empty snail shells into the water and watching what happened. One new shell triggers a stampede. The biggest crab claims it. The second crab takes the shell the first one just abandoned. And so on down the size gradient, everyone upgrading one slot. The average chain moves about 2.5 to 3 crabs into better housing per new shell. Chase’s writeup calls it a vacancy chain and notes, with a straight face, that the same mathematics had already been run on Manhattan apartments. That part is real. Kristof, doing housing research for New York City in the 1960s, watched every new apartment trigger a moving chain of upgrades through the existing stock. White wrote the housing version up himself in 1971.
Shells, apartments, parishes, seats. One mechanism, running everything from tide pools to Manhattan real estate. I find this genuinely delightful. Nature discovered recruiting before we did, and the crabs have less paperwork.
The arithmetic
Since 2000, the Bureau of Labor Statistics has kept a census of exactly this machine. The Job Openings and Labor Turnover Survey counts, every month, how many Americans were hired, how many quit, and how many were laid off. I pulled the whole series and did the math, and the math rearranged how I think about my own industry.
From January 2001 through August 2026: about 1.60 billion hires. Over the same window, payroll employment grew by about 26 million jobs. Divide one by the other.
Under 2% of American hiring creates anything.
Round numbers for the people who like round numbers: about 247,000 people started a job every working day of that quarter century. Of those, about 4,000 a day landed on a seat that did not exist the month before. The other 243,000 walked into a vacancy that a resignation, a promotion, or a layoff had just opened. A job opening is, with almost no exceptions, somebody else’s departure wearing a nice title.
And the machine never speeds up. That’s the second surprise. Monthly hires stayed inside a 3.6M to 8.1M band through the dot-com crash, the financial crisis, the COVID whiplash, and the AI boom. In 294 of those 308 months, hires outnumbered quits and layoffs combined. The 14 exceptions are the interesting ones: 2001, the fall of 2008 through mid-2009, and March and April 2020. April 2020 is the month the machine almost stopped: 1.85M hires against 5M exits. Two months later it spun faster than it ever has, 8.1M hires, and even then, 98% of it was seats changing hands. The drum doesn’t speed up in good times or slow down in bad ones. It spins, and it only changes direction when something breaks.
“But people my age change jobs constantly,” you may be thinking, and the data agrees with you in the strangest way. Quits and layoffs remove about 36% of payroll employment every year, yet the median job still lasts over four years. Both facts are true because the churn lands almost entirely on the young. The BLS followed one whole birth cohort from 1957 to age 58: 12.9 jobs per person, and over 40% of them before age 25. Your grandparents’ generation job-hopped too. They just did it at an age when nobody was writing LinkedIn posts about it.
So when someone tells me they’re “looking for a job,” I now hear the sentence differently. The literal object barely exists. What they’re looking for is a turn.
The quiet reason
Some hiring does create seats. Why is that share so small, and why does nobody talk about it?
Because the loud reason and the quiet reason for a job opening get confused, and only one of them reaches the job boards.
The loud reason is replacement. Someone left, the seat is warm, the work is piling up, and the requisition is born as an emergency. This is the 98%.
The quiet reason is growth. New seats come from new revenue lines, new funding, new technology. And companies staff those quietly, internally, first: reassignments, transfers, people who were already there getting the scope. A brand new seat only reaches the external market after the internal options ran out. The job board, the thing everyone stares at, is where chains go after they’ve already been shopping indoors.
Bidwell’s Paying More to Get Less quantifies what the outdoor market gets you, using personnel data from an investment bank. External hires into similar roles were paid about 18% more, performed worse for two years, and quit at higher rates than people promoted into the same seats. The company paid an outside premium, at auction, for the privilege of a slower start. If you’ve ever wondered why external recruiting is expensive and mildly demoralizing for everyone involved, that 18% is a decent place to start.
What the chain sees
Follow any opening backward and you stop counting candidates and start counting people who already hold seats. Three things about that chain do most of the damage, and all three are funny once you stop taking them personally.
First: hiring follows exits. When quits peaked at 4.5M in April 2022, hires peaked the same month. Over 25 years, monthly hires and monthly quits move almost in lockstep. We like to think hiring responds to opportunity. It responds to other people leaving.
Second: the chain runs on availability, not skill. In Chase’s tide pool, chains ended early when the initiating crab sat in a damaged shell, because damaged vacancies stayed empty. Every crab in that pool was perfectly qualified to live in a good shell. What mattered was who happened to be positioned when the shell opened. Kroft, Lange and Notowidigdo sent about 12,000 fictitious resumes to real postings in 100 US cities and found that callback rates fell from roughly 7% to 4% within eight months of unemployment, a 45% collapse, sharpest in the tightest markets. Nothing about the candidates changed. Their position in the chain changed. The same paper trail runs through entire cohorts: Oreopoulos, von Wachter and Heisz found that graduating into a downturn suppressed earnings for a decade, and Kahn found the same for the US college class of the early 1980s. The pool is not empty. The pool is waiting for a shell.
Third: the chain is blind, and it recruits the blind. Every observer in it is staring at the last move. This is why no-poach agreements work so disturbingly well. Silicon Valley, an industry whose entire mythology is about mobility, spent years with a formal pact not to cold call each other’s engineers. The Justice Department made them stop in 2010, and the private class action settled for $415 million in 2015. You can strangle a whole rotation without deleting a single job. The everyday version of the same reflex is the 70% of the workforce that LinkedIn’s own surveys call passive, the people not actively job hunting, whose availability is known only to their networks. Rajkumar and colleagues ran a five-year randomized experiment on 20 million LinkedIn members and confirmed Granovetter causally: weak ties move jobs, strongest ties move them least. Openings travel through chains, and chains charge rent to outsiders.
Want to watch the blindness in real time? Watch recruiters converge on a candidate. Say the candidate’s true odds of accepting an offer are 35%, and each recruiter’s read is a bit better than a coin flip. One independent read moves you to 56%. A second independent read takes you to 75%. A third should take you higher. But the third recruiter has usually already heard the first two opinions. Once two reads agree, the third confirms what everyone believes and carries no new information. The estimate stops at 75% and never moves again, no matter how many people “confirm” it. Every candidate who lost an offer to a panel that all thought alike has met this room. The only reference check worth anything is the one made blind, and nobody runs it that way, for reasons that made sense the day the process was designed and never again after.
One boundary on all of this, stated where it belongs: the crab thing is a shape I keep seeing, not a mechanism I’ve measured in labor data. Crabs trade up in minutes; companies take months and insert interviews, politics, and counteroffers into every hop. What transfers is the constraint. Discrete seats. Occupancy rules. Nothing more.
But the drum itself is the part that survives contact with the data. Openings are other people’s departures; growth is the rounding error. If you want to know where the jobs are, stop watching the job. Watch who just left.
The power plant
I’ve been trying to find the right name for what hiring actually is, because “recruiting” describes the choreography and misses the physics.
Here’s the closest I’ve found. Some power stations sell their waste heat to nearby buildings. It’s called cogeneration: the useful output is a byproduct of a process everyone was already running for other reasons. Hiring is cogeneration. Seats are the waste heat of people leaving seats. Nobody plans the heat. It shows up because the turbine is spinning.
Once you see it that way, three things you’d never guess become obvious.
Your own churn is your hiring forecast. At 36% annual turnover, most of next year’s requisitions already exist; they’re just waiting on exit dates. A company that models its attrition knows its hiring volume a year out, and the rest of the industry is pretending to be surprised by Dave.
Your neighbor’s churn is your pipeline. The drum is economy-wide. Every resignation across the street frees up a seat somewhere, which is why hiring surges cluster in time, and why the poaching bans “worked”: they suppressed the inflow at the source, at the cost of $415 million and a decade of suppressed wages.
And internal mobility is your chain length. If one departure digs deep, if the seat rolls through three people before it leaves the building, your external funnel is a formality. If every opening exits immediately to an outside hire, you’re paying the 18% every time, and calling it strategy.
If this frame is wrong, this is how: the drum’s share collapses. The 98% is a property of an economy that grows by reallocating, and allocation can only recycle so much. If AI adoption creates genuinely new seats the way electrification once did, hiring would run ahead of exits for years at a stretch. I would genuinely enjoy being wrong this way. The payroll numbers can’t see any of this either, they only show the net residue, which is why the mechanism stayed invisible to me for five years while I worked inside it. The count has a small edge too: a replaced seat that was also eliminated and recreated counts once in each direction, so the true figure is “under 3%” rather than 2%. The conclusion survives either rounding.
Openings are other people’s departures. Growth is the rounding error. Stop watching the job. Watch who just left.
And if you’re in the pool, waiting: the shell you’re waiting for is attached to somebody. That is not a metaphor. It is the mechanism.